def process_file(self, mel, speaker_index, speaker_index_2, notes, sess):
datasets = "".join("_" + x.lower() for x in config.datasets)
with h5py.File(config.stat_file, mode='r') as stat_file:
max_feat = stat_file["feats_maximus"][()] + 0.001
min_feat = stat_file["feats_minimus"][()] - 0.001
mel = (mel - min_feat) / (max_feat - min_feat)
notes = notes / np.round(max_feat[-2])
in_batches_mel, nchunks_in = utils.generate_overlapadd(mel)
in_batches_notes, nchunks_in = utils.generate_overlapadd(notes)
out_batches_mel = []
out_batches_f0 = []
out_batches_vuv = []
for in_batch_mel, in_batch_notes in zip(in_batches_mel,
in_batches_notes):
speaker = np.repeat(speaker_index, config.batch_size)
speaker_2 = np.repeat(speaker_index_2, config.batch_size)
feed_dict = {self.input_placeholder: in_batch_mel, self.speaker_labels:speaker,self.speaker_labels_1:speaker_2,\
self.notes_placeholder:in_batch_notes, self.is_train: False}
mel, f0, vuv = sess.run([self.output, self.f0, self.vuv],
feed_dict=feed_dict)
out_batches_mel.append(mel)
out_batches_f0.append(f0)
out_batches_vuv.append(vuv)
out_batches_mel = np.array(out_batches_mel)
out_batches_f0 = np.array(out_batches_f0)
out_batches_vuv = np.array(out_batches_vuv)
out_batches_mel = utils.overlapadd(out_batches_mel, nchunks_in)
out_batches_f0 = utils.overlapadd(out_batches_f0, nchunks_in)
out_batches_vuv = utils.overlapadd(out_batches_vuv, nchunks_in)
out_batches_mel = out_batches_mel[:, :-2] * (
max_feat[:-2] - min_feat[:-2]) + min_feat[:-2]
out_batches_f0 = np.clip(out_batches_f0, 0.0, 1.0) * (
max_feat[-2] - min_feat[-2]) + min_feat[-2]
out_batches_vuv = out_batches_vuv * (max_feat[-1] -
min_feat[-1]) + min_feat[-1]
out_batches_vuv = np.round(out_batches_vuv)
return out_batches_mel, out_batches_f0, out_batches_vuv
def process_file(self, mel, speaker_index, speaker_index_2, sess):
datasets = "".join("_"+x.lower() for x in config.datasets)
with h5py.File(config.stat_file, mode='r') as stat_file:
max_feat = stat_file["feats_maximus"][()] + 0.001
min_feat = stat_file["feats_minimus"][()] - 0.001
mel = (mel - min_feat)/(max_feat-min_feat)
in_batches_mel, nchunks_in = utils.generate_overlapadd(mel)
out_batches_mel = []
for in_batch_mel in in_batches_mel :
speaker = np.repeat(np.expand_dims(speaker_index,0),config.batch_size, axis=0)
speaker_2 = np.repeat(np.expand_dims(speaker_index_2,0),config.batch_size, axis=0)
feed_dict = {self.input_placeholder: in_batch_mel[:,:,:-2], self.speaker_labels:speaker,self.speaker_labels_1:speaker_2, self.is_train: False}
mel = sess.run(self.output, feed_dict=feed_dict)
out_batches_mel.append(mel)
out_batches_mel = np.array(out_batches_mel)
out_batches_mel = utils.overlapadd(out_batches_mel,nchunks_in)
out_batches_mel = out_batches_mel*(max_feat[:-2] - min_feat[:-2]) + min_feat[:-2]
return out_batches_mel
def extract_feature(self, mel, sess):
datasets = "".join("_"+x.lower() for x in config.datasets)
mel = np.clip(mel, 0.0, 1.0)
in_batches_mel, nchunks_in = utils.generate_overlapadd(mel)
out_batches_mel = []
for in_batch_mel in in_batches_mel :
feed_dict = {self.stft_placeholder: in_batch_mel, self.is_train: False}
mel = sess.run(self.content_embedding_stft, feed_dict=feed_dict)
out_batches_mel.append(mel)
out_batches_mel = np.array(out_batches_mel)
out_batches_mel = utils.overlapadd(out_batches_mel,nchunks_in)
return out_batches_mel
def process_file(self, mel, sess):
datasets = "".join("_"+x.lower() for x in config.datasets)
with h5py.File(config.stat_file, mode='r') as stat_file:
max_feat = stat_file["feats_maximus"][()] + 0.001
min_feat = stat_file["feats_minimus"][()] - 0.001
mel = np.clip(mel, 0.0, 1.0)
in_batches_mel, nchunks_in = utils.generate_overlapadd(mel)
out_batches_mel = []
out_batches_f0 = []
out_batches_vuv = []
for in_batch_mel in in_batches_mel :
feed_dict = {self.stft_placeholder: in_batch_mel, self.stft_placeholder_1: in_batch_mel, self.is_train: False}
mel, f0, vuv = sess.run([self.output_stft, self.f0, self.vuv], feed_dict=feed_dict)
out_batches_mel.append(mel)
out_batches_f0.append(f0)
out_batches_vuv.append(vuv)
out_batches_mel = np.array(out_batches_mel)
out_batches_f0 = np.array(out_batches_f0)
out_batches_vuv = np.array(out_batches_vuv)
out_batches_mel = utils.overlapadd(out_batches_mel,nchunks_in)
out_batches_f0 = utils.overlapadd(out_batches_f0,nchunks_in)
out_batches_vuv = utils.overlapadd(out_batches_vuv,nchunks_in)
out_batches_mel = out_batches_mel*(max_feat[:-2] - min_feat[:-2]) + min_feat[:-2]
out_batches_f0 = out_batches_f0*(max_feat[-2] - min_feat[-2]) + min_feat[-2]
out_batches_vuv = out_batches_vuv*(max_feat[-1] - min_feat[-1]) + min_feat[-1]
out_batches_vuv = np.round(out_batches_vuv)
return out_batches_mel, out_batches_f0, out_batches_vuv